OpenAI-compatible custom base_url strips openai/ from Featherless model IDs

[agentcypot]

Description

When configuring Featherless as an OpenAI-compatible provider, OpenFang appears to send gpt-oss-120b to the provider instead of the exact configured model ID openai/gpt-oss-120b.

Featherless requires the full namespaced model ID, so the request fails with:

Error: Model not found: {"error":{"message":"The model `gpt-oss-120b` does not exist.","type":"invalid_request_error","code":"model_not_found"}}

Configuration:

[default_model]
provider = "openai"
model = "openai/gpt-oss-120b"
api_key_env = "FEATHERLESS_API_KEY"
base_url = "https://api.featherless.ai/v1"

Confirmed workaround:

Using a Featherless model ID that does not start with openai/, such as zai-org/GLM-5.1, works with the same provider, base_url, and api_key_env settings.

This may be related to #856, which also involved OpenFang rewriting provider-specific model IDs. This report is specifically about provider = "openai" with a custom base_url, where model IDs should be treated as opaque strings and passed through exactly.

Expected Behavior

For OpenAI-compatible providers with a custom base_url, OpenFang should preserve the configured model string exactly, including slashes.

The value sent to the provider should be:

openai/gpt-oss-120b

not:

gpt-oss-120b

Steps to Reproduce

1. Configure OpenFang with Featherless as an OpenAI-compatible provider:

[default_model]
provider = "openai"
model = "openai/gpt-oss-120b"
api_key_env = "FEATHERLESS_API_KEY"
base_url = "https://api.featherless.ai/v1"

2. Start OpenFang.

3. Send any request that uses the default model.

4. Observe that Featherless returns a model-not-found error for gpt-oss-120b.

5. Change only the model value to a Featherless model ID that does not start with openai/, for example:

model = "zai-org/GLM-5.1"

6. Repeat the request.

7. Observe that the request works with the same provider, base_url, and api_key_env settings.

OpenFang Version

0.6.9

Operating System

Linux (x86_64)

Logs / Screenshots

The relevant error is:

Error: Model not found: {"error":{"message":"The model `gpt-oss-120b` does not exist.","type":"invalid_request_error","code":"model_not_found"}}

The configured model was:

openai/gpt-oss-120b

The provider error indicates the request was made with:

gpt-oss-120b